https://openai.com/blog/introducing-gpts

Introducing GPTs

You can now create custom versions of ChatGPT that combine instructions, extra knowledge, and any combination of skills.

November 6, 2023

We’re rolling out custom versions of ChatGPT that you can create for a specific purpose—called GPTs. GPTs are a new way for anyone to create a tailored version of ChatGPT to be more helpful in their daily life, at specific tasks, at work, or at home—and then share that creation with others. For example, GPTs can help you learn the rules to any board game, help teach your kids math, or design stickers.

특정 목적을 위해 생성할 수 있는 ChatGPT의 사용자 지정 버전(GPTs)이 출시됩니다. GPTs는 누구나 일상 생활, 특정 작업, 직장 또는 집에서 더 도움이 되도록 맞춤형 버전의 ChatGPT를 만들고 해당 창작물을 다른 사람들과 공유할 수 있는 새로운 방법입니다. 예를 들어, GPTs는 보드 게임의 규칙을 배우거나, 자녀에게 수학을 가르치거나, 스티커를 디자인하는 데 도움이 될 수 있습니다.

Anyone can easily build their own GPT—no coding is required. You can make them for yourself, just for your company’s internal use, or for everyone. Creating one is as easy as starting a conversation, giving it instructions and extra knowledge, and picking what it can do, like searching the web, making images or analyzing data. Try it out at chat.openai.com/create.

누구나 쉽게 자신만의 GPT를 구축할 수 있습니다. 코딩이 필요하지 않습니다. 회사 내부용으로만 사용할 수도 있고 모든 사람을 위해 만들 수도 있습니다. 대화를 시작하고, 지침과 추가 지식을 제공하고, 웹 검색, 이미지 만들기, 데이터 분석 등 할 수 있는 작업을 선택하는 것만큼 쉽습니다. chat.openai.com/create에서 사용해 보세요.

Example GPTs are available today for ChatGPT Plus and Enterprise users to try out including Canva and Zapier AI Actions. We plan to offer GPTs to more users soon.

ChatGPT Plus 및 Enterprise 사용자는 오늘부터 Canva 및 Zapier AI 작업을 포함한 예제 GPTs를 사용해 볼 수 있습니다. 우리는 곧 더 많은 사용자에게 GPTs를 제공할 계획입니다.

Learn more about our OpenAI DevDay announcements for new models and developer products.

새로운 모델 및 개발자 제품에 대한 것들은 OpenAI DevDay 에서의 발표에 대해 자세히 알아보세요.

오늘부터 GPTs를 생성하고 공개적으로 공유할 수 있습니다. 이번 달 말에는 검증된 제작자의 창작물을 선보이는 GPT 스토어를 출시할 예정입니다. 매장에 들어가면 GPTs를 검색할 수 있으며 순위표에 오를 수 있습니다. 또한 생산성, 교육, '재미를 위한' 카테고리에서 가장 유용하고 즐거운 GPTs를 집중 조명할 것입니다. 앞으로 몇 달 안에 GPT를 사용하는 사람 수에 따라 수익을 얻을 수도 있습니다.

GPTs는 계속해서 더욱 유용하고 스마트해질 것이며, 결국 GPT가 현실 세계에서 실제 작업을 수행하도록 할 수 있게 될 것입니다. AI 분야에서 이러한 시스템은 종종 "에이전트"로 논의됩니다. 우리는 이 미래를 향해 점진적으로 나아가는 것이 중요하다고 생각합니다. 왜냐하면 세심한 기술 및 안전 작업과 사회가 적응할 시간이 필요하기 때문입니다. 우리는 사회적 영향에 대해 깊이 생각해 왔으며 곧 더 많은 분석을 공유할 예정입니다.

내장된 기능을 사용하는 것 외에도 GPTs에서 하나 이상의 API를 사용할 수 있도록 하여 맞춤 작업을 정의할 수도 있습니다. 플러그인과 마찬가지로 작업을 통해 GPTs는 외부 데이터를 통합하거나 실제 세계와 상호 작용할 수 있습니다. GPTs를 데이터베이스에 연결하거나, 이메일에 연결하거나, 쇼핑 도우미로 활용하세요. 예를 들어 여행 목록 데이터베이스를 통합하거나, 사용자의 이메일 받은 편지함을 연결하거나, 전자 상거래 주문을 촉진할 수 있습니다.

기업은 수요일부터 GPTs를 시작할 수 있습니다. 이제 회사 내부 사용자가 코드 없이 내부 전용 GPTs를 설계하고 작업공간에 안전하게 게시할 수 있는 권한을 부여할 수 있습니다. 관리 콘솔을 사용하면 GPTs 공유 방법과 외부 GPTs를 비즈니스 내에서 사용할 수 있는지 여부를 선택할 수 있습니다. ChatGPT Enterprise의 모든 사용과 마찬가지로 우리는 모델을 개선하기 위해 GPTs와의 대화를 사용하지 않습니다.

https://openai.com/blog/new-models-and-developer-products-announced-at-devday

New models and developer products announced at DevDay

GPT-4 Turbo with 128K context and lower prices, the new Assistants API, GPT-4 Turbo with Vision, DALL·E 3 API, and more.

November 6, 2023

Today, we shared dozens of new additions and improvements, and reduced pricing across many parts of our platform. These include:

오늘 우리는 플랫폼의 여러 부분에 걸쳐 수십 가지의 새로운 추가 및 개선 사항과 가격 인하를 공유했습니다. 여기에는 다음이 포함됩니다.

• New GPT-4 Turbo model that is more capable, cheaper and supports a 128K context window
• 더 유능하고 저렴하며 128K 컨텍스트 창을 지원하는 새로운 GPT-4 Turbo 모델
• New Assistants API that makes it easier for developers to build their own assistive AI apps that have goals and can call models and tools
• 개발자가 목표가 있고 모델과 도구를 호출할 수 있는 자체 보조 AI 앱을 더 쉽게 구축할 수 있게 해주는 새로운 Assistant API
• New multimodal capabilities in the platform, including vision, image creation (DALL·E 3), and text-to-speech (TTS)
• 비전, 이미지 생성(DALL·E 3) 및 TTS(텍스트 음성 변환)를 포함한 플랫폼의 새로운 다중 모드 기능

We’ll begin rolling out new features to OpenAI customers starting at 1pm PT today.

오늘 오후 1시(태평양 표준시)부터 OpenAI 고객에게 새로운 기능을 선보일 예정입니다.

Learn more about OpenAI DevDay announcements for ChatGPT.

ChatGPT에 대한 OpenAI DevDay 공지사항에 대해 자세히 알아보세요.

New modalities in the API

GPT-4 Turbo는 이미지를 Chat Completions API의 입력으로 받아들여 캡션 생성, 실제 이미지 세부 분석, 그림이 포함된 문서 읽기 등의 사용 사례를 지원합니다. 예를 들어, BeMyEyes는 이 기술을 사용하여 시각 장애가 있거나 시력이 낮은 사람들이 제품 식별이나 매장 탐색과 같은 일상 업무를 수행할 수 있도록 돕습니다. 개발자는 API에서 gpt-4-vision-preview를 사용하여 이 기능에 액세스할 수 있습니다. 우리는 안정적인 릴리스의 일부로 주요 GPT-4 Turbo 모델에 비전 지원을 출시할 계획입니다. 가격은 입력 이미지 크기에 따라 다릅니다. 예를 들어 1080×1080 픽셀의 이미지를 GPT-4 Turbo로 전달하는 데 드는 비용은 $0.00765입니다. 비전 가이드를 확인해 보세요.

개발자는 dall-e-3를 모델로 지정하여 Images API를 통해 최근 ChatGPT Plus 및 Enterprise 사용자에게 출시된 DALL·E 3를 앱과 제품에 직접 통합할 수 있습니다. Snap, Coca-Cola, Shutterstock과 같은 회사에서는 DALL·E 3를 사용하여 고객과 캠페인을 위한 이미지와 디자인을 프로그래밍 방식으로 생성했습니다. 이전 버전의 DALL·E와 마찬가지로 API에는 개발자가 응용 프로그램을 오용으로부터 보호할 수 있도록 조정 기능이 내장되어 있습니다. 우리는 생성된 이미지당 $0.04부터 시작하는 가격으로 다양한 형식과 품질 옵션을 제공합니다. API에서 DALL·E 3을 시작하는 방법에 대한 가이드를 확인하세요.

미세 조정이 제공할 수 있는 것보다 훨씬 더 많은 사용자 정의가 필요한 조직을 위해(특히 극도로 큰 독점 데이터 세트(최소 수십억 개의 토큰)가 있는 도메인에 적용 가능)를 위해 우리는 또한 사용자 정의 모델 프로그램을 출시하여 일부 조직에 특정 도메인에 맞게 맞춤형 GPT-4를 교육하는 OpenAI 연구원 전용 그룹입니다. 여기에는 추가 도메인별 사전 교육 수행부터 특정 도메인에 맞춰진 사용자 정의 RL 사후 교육 프로세스 실행까지 모델 교육 프로세스의 모든 단계를 수정하는 것이 포함됩니다. 조직은 맞춤형 모델에 독점적으로 액세스할 수 있습니다. 기존 기업 개인 정보 보호 정책에 따라 사용자 지정 모델은 다른 고객에게 제공 또는 공유되지 않으며 다른 모델을 교육하는 데 사용되지 않습니다. 또한 맞춤형 모델을 훈련하기 위해 OpenAI에 제공된 독점 데이터는 다른 어떤 맥락에서도 재사용되지 않습니다. 이는 시작하기에 매우 제한적이고 비용이 많이 드는 프로그램입니다. 관심 있는 조직은 여기에서 신청할 수 있습니다.

애플리케이션 확장을 돕기 위해 모든 유료 GPT-4 고객의 분당 토큰 한도를 두 배로 늘립니다. 비율 제한 페이지에서 새로운 비율 제한을 볼 수 있습니다. 또한 자동 요금 한도 증가를 결정하는 사용량 계층을 게시했으므로 사용량 한도가 자동으로 확장되는 방식에 대해 예상할 수 있습니다. 이제 계정 설정에서 사용 한도 증가를 요청할 수 있습니다.

ChatGPT에 대한 OpenAI DevDay 공지사항에 대해 자세히 알아보세요.

GPT-4 API general availability and deprecation of older models in the Completions API (openai.com)

GPT-4 API general availability and deprecation of older models in the Completions API

GPT-3.5 Turbo, DALL·E and Whisper APIs are also generally available, and we are releasing a deprecation plan for older models of the Completions API, which will retire at the beginning of 2024.

GPT-3.5 Turbo, DALL·E 및 Whisper API도 generally 사용할 수 있으며, Completions API의 이전 모델에 대한 지원 중단 계획을 발표합니다. 이 모델들은 2024년 초에 사용 중지됩니다.

July 6, 2023

Starting today, all paying API customers have access to GPT-4. In March, we introduced the ChatGPT API, and earlier this month we released our first updates to the chat-based models. We envision a future where chat-based models can support any use case. Today we’re announcing a deprecation plan for older models of the Completions API, and recommend that users adopt the Chat Completions API.

오늘부터 모든 유료 API 고객은 GPT-4에 액세스할 수 있습니다. 3월에 우리는 ChatGPT API를 도입했고 이달 초에는 채팅 기반 모델에 대한 첫 번째 업데이트를 발표했습니다. 우리는 채팅 기반 모델이 모든 사용 사례를 지원할 수 있는 미래를 상상합니다. 오늘 우리는 Completions API의 이전 모델에 대한 지원 중단 계획을 발표하고 사용자가 Chat Completions API를 채택할 것을 권장합니다.

GPT-4 API general availability

GPT-4 is our most capable model. Millions of developers have requested access to the GPT-4 API since March, and the range of innovative products leveraging GPT-4 is growing every day. Today all existing API developers with a history of successful payments can access the GPT-4 API with 8K context. We plan to open up access to new developers by the end of this month, and then start raising rate-limits after that depending on compute availability.

GPT-4는 가장 capable한 모델입니다. 3월부터 수백만 명의 개발자가 GPT-4 API에 대한 액세스를 요청했으며 GPT-4를 활용하는 혁신적인 제품의 범위가 매일 증가하고 있습니다. 현재 성공적인 결제 기록이 있는 모든 기존 API 개발자는 8K 컨텍스트로 GPT-4 API에 액세스할 수 있습니다. 이달 말까지 새로운 개발자에게 액세스를 허용한 다음 컴퓨팅 가용성에 따라 속도 제한을 높일 계획입니다.

Based on the stability and readiness of these models for production-scale use, we are also making the GPT-3.5 Turbo, DALL·E and Whisper APIs generally available. We are working on safely enabling fine-tuning for GPT-4 and GPT-3.5 Turbo and expect this feature to be available later this year.

생산 규모 사용을 위한 이러한 모델의 안정성과 준비성을 기반으로 GPT-3.5 Turbo, DALL·E 및 Whisper API를 generally 사용할 수 있도록 만들고 있습니다. 우리는 GPT-4 및 GPT-3.5 Turbo에 대한 fine-tuning을 안전하게 활성화하기 위해 노력하고 있으며 이 기능은 올해 말에 제공될 예정입니다.

Moving from text completions to chat completions

We introduced the Chat Completions API in March, and it now accounts for 97% of our API GPT usage.

3월에 Chat Completions API를 도입했으며 현재 API GPT 사용량의 97%를 차지합니다.

The initial Completions API was introduced in June 2020 to provide a freeform text prompt for interacting with our language models. We’ve since learned that we can often provide better results with a more structured prompt interface. The chat-based paradigm has proven to be powerful, handling the vast majority of previous use cases and new conversational needs, while providing higher flexibility and specificity. In particular, the Chat Completions API’s structured interface (e.g., system messages, function calling) and multi-turn conversation capabilities enable developers to build conversational experiences and a broad range of completion tasks. It also helps lower the risk of prompt injection attacks, since user-provided content can be structurally separated from instructions.

초기 Completions API는 2020년 6월에 도입되어 언어 모델과 상호 작용하기 위한 자유 형식 텍스트 프롬프트를 제공합니다. 이후 우리는 보다 구조화된 프롬프트 인터페이스를 통해 종종 더 나은 결과를 제공할 수 있다는 것을 알게 되었습니다. 채팅 기반 패러다임은 이전 사용 사례와 새로운 대화 요구 사항의 대부분을 처리하는 동시에 더 높은 유연성과 특수성을 제공하는 강력한 것으로 입증되었습니다. 특히 Chat Completions API의 구조화된 인터페이스(예: 시스템 메시지, 함수 호출) 및 멀티턴 대화 기능을 통해 개발자는 대화형 경험과 광범위한 완료 작업을 구축할 수 있습니다. 또한 사용자가 제공한 콘텐츠와 지침을 구조적으로 분리할 수 있으므로 프롬프트 인젝션 공격의 위험을 줄이는 데 도움이 됩니다.

We plan to continue investing most of our platform efforts in this direction, as we believe it will offer an increasingly capable and easy-to-use experience for developers. We’re working on closing the last few remaining gaps of the Chat Completions API quickly, such as log probabilities for completion tokens and increased steerability to reduce the “chattiness” of responses.

우리는 이 방향으로 대부분의 플랫폼 노력을 계속해서 투자할 계획입니다. 이는 개발자에게 점점 더 기능이 풍부하고 사용하기 쉬운 경험을 제공할 것이라고 믿기 때문입니다. 완료 토큰에 대한 로그 확률 및 응답의 "채팅성"을 줄이기 위한 향상된 조종성 등 Chat Completions API의 마지막 몇 가지 남은 격차를 신속하게 해결하기 위해 노력하고 있습니다.

Deprecation of older models in the Completions API

As part of our increased investment in the Chat Completions API and our efforts to optimize our compute capacity, in 6 months we will be retiring some of our older models using the Completions API. While this API will remain accessible, we will label it as “legacy” in our developer documentation starting today. We plan for future model and product improvements to focus on the Chat Completions API, and do not have plans to publicly release new models using the Completions API.

Chat Completions API에 대한 투자 증가와 컴퓨팅 용량을 최적화하기 위한 노력의 일환으로 6개월 후에 Completions API를 사용하는 이전 모델 중 일부를 폐기할 예정입니다. 이 API는 계속 액세스할 수 있지만 오늘부터 개발자 문서에서 "레거시"로 표시됩니다. Chat Completions API에 집중하기 위해 향후 모델 및 제품 개선을 계획하고 Completions API를 사용하여 새 모델을 공개적으로 출시할 계획은 없습니다.

Starting January 4, 2024, older completion models will no longer be available, and will be replaced with the following models:

2024년 1월 4일부터 이전 완료 모델을 더 이상 사용할 수 없으며 다음 모델로 대체됩니다.

Applications using the stable model names for base GPT-3 models (ada, babbage, curie, davinci) will automatically be upgraded to the new models listed above on January 4, 2024. The new models will also be accessible in the coming weeks for early testing by specifying the following model names in API calls: ada-002, babbage-002, curie-002, davinci-002.

기본 GPT-3 모델(ada, babbage, curie, davinci)에 안정적인 모델 이름을 사용하는 애플리케이션은 2024년 1월 4일에 위에 나열된 새 모델로 자동 업그레이드됩니다. API 호출에 다음 모델 이름을 지정하여 테스트: ada-002, babbage-002, curie-002, davinci-002.

Developers using other older completion models (such as text-davinci-003) will need to manually upgrade their integration by January 4, 2024 by specifying gpt-3.5-turbo-instruct in the “model” parameter of their API requests. gpt-3.5-turbo-instruct is an InstructGPT-style model, trained similarly to text-davinci-003. This new model is a drop-in replacement in the Completions API and will be available in the coming weeks for early testing.

다른 이전 완성 모델(예: text-davinci-003)을 사용하는 개발자는 API 요청의 "모델" 매개변수에 gpt-3.5-turbo-instruct를 지정하여 2024년 1월 4일까지 통합을 수동으로 업그레이드해야 합니다. gpt-3.5-turbo-instruct는 text-davinci-003과 유사하게 훈련된 InstructGPT 스타일 모델입니다. 이 새 모델은 Completions API의 드롭인 대체품이며 초기 테스트를 위해 앞으로 몇 주 안에 사용할 수 있습니다.

Developers wishing to continue using their fine-tuned models beyond January 4, 2024 will need to fine-tune replacements atop the new base GPT-3 models (ada-002, babbage-002, curie-002, davinci-002), or newer models (gpt-3.5-turbo, gpt-4). Once this feature is available later this year, we will give priority access to GPT-3.5 Turbo and GPT-4 fine-tuning to users who previously fine-tuned older models. We acknowledge that migrating off of models that are fine-tuned on your own data is challenging. We will be providing support to users who previously fine-tuned models to make this transition as smooth as possible.

 fine-tuned된 모델을 2024년 1월 4일 이후에도 계속 사용하려는 개발자는 새로운 기본 GPT-3 모델(ada-002, babbage-002, curie-002, davinci-002) 이나 최근 모델(gpt-3.5-turbo, gpt-4)에서  fine-tuning해야 합니다 . 올해 후반에 이 기능을 사용할 수 있게 되면 이전 모델을 fine-tuning한 사용자에게 GPT-3.5 Turbo 및 GPT-4 fine-tuning에 대한 우선 액세스 권한을 부여할 것입니다. 자체 데이터에 대해 미세 조정된 모델에서 마이그레이션하는 것이 어렵다는 것을 알고 있습니다. 이전에 모델을 미세 조정한 사용자에게 이러한 전환이 최대한 원활하게 이루어지도록 지원을 제공할 예정입니다.

In the coming weeks, we will reach out to developers who have recently used these older models, and will provide more information once the new completion models are ready for early testing.

앞으로 몇 주 동안 이러한 이전 모델을 최근에 사용한 개발자에게 연락을 취하고 새로운 완성 모델이 초기 테스트 준비가 되면 더 많은 정보를 제공할 것입니다.

Deprecation of older embeddings models

Users of older embeddings models (e.g., text-search-davinci-doc-001) will need to migrate to text-embedding-ada-002 by January 4, 2024. We released text-embedding-ada-002 in December 2022, and have found it more capable and cost effective than previous models. Today text-embedding-ada-002 accounts for 99.9% of all embedding API usage.

이전 임베딩 모델(예: text-search-davinci-doc-001) 사용자는 2024년 1월 4일까지 text-embedding-ada-002로 마이그레이션해야 합니다. 2022년 12월에 text-embedding-ada-002를 출시했으며 이전 모델보다 성능이 뛰어나고 비용 효율적이라는 사실을 알게 되었습니다. 현재 text-embedding-ada-002는 모든 임베딩 API 사용량의 99.9%를 차지합니다.

We recognize this is a significant change for developers using those older models. Winding down these models is not a decision we are making lightly. We will cover the financial cost of users re-embedding content with these new models. We will be in touch with impacted users over the coming days.

우리는 이것이 이전 모델을 사용하는 개발자에게 중요한 변화라는 것을 알고 있습니다. 이러한 모델을 종료하는 것은 우리가 가볍게 내리는 결정이 아닙니다. 우리는 사용자가 이러한 새 모델로 콘텐츠를 다시 임베딩하는 재정적 비용을 충당할 것입니다. 앞으로 며칠 동안 영향을 받는 사용자에게 연락을 드릴 것입니다.

Deprecation of the Edits API

Users of the Edits API and its associated models (e.g., text-davinci-edit-001 or code-davinci-edit-001) will need to migrate to GPT-3.5 Turbo by January 4, 2024. The Edits API beta was an early exploratory API, meant to enable developers to return an edited version of the prompt based on instructions. We took the feedback from the Edits API into account when developing gpt-3.5-turbo and the Chat Completions API, which can now be used for the same purpose:

Edits API 및 관련 모델(예: text-davinci-edit-001 또는 code-davinci-edit-001) 사용자는 2024년 1월 4일까지 GPT-3.5 Turbo로 마이그레이션해야 합니다. Edits API 베타는 초기 버전이었습니다. 개발자가 지침에 따라 프롬프트의 편집된 버전을 반환할 수 있도록 하는 탐색 API. 우리는 gpt-3.5-turbo 및 Chat Completions API를 개발할 때 Edits API의 피드백을 고려했으며 이제 동일한 용도로 사용할 수 있습니다.

https://openai.com/blog/function-calling-and-other-api-updates

Function calling and other API updates

Announcements, Product

We released gpt-3.5-turbo and gpt-4 earlier this year, and in only a short few months, have seen incredible applications built by developers on top of these models.

우리는 올해 초에 gpt-3.5-turbo 및 gpt-4를 출시했고 불과 몇 달 만에 개발자들이 이러한 모델 위에 구축한 놀라운 애플리케이션을 보았습니다.

Today, we’re following up with some exciting updates:

오늘 우리는 다음과 같은 몇 가지 흥미로운 업데이트를 진행합니다.

• new function calling capability in the Chat Completions API
• Chat Completions API의 새로운 함수 호출 기능
• updated and more steerable versions of gpt-4 and gpt-3.5-turbo
• gpt-4 및 gpt-3.5-turbo의 업데이트되고 조정 가능한 버전
• new 16k context version of gpt-3.5-turbo (vs the standard 4k version)
• gpt-3.5-turbo의 새로운 16k 컨텍스트 버전(표준 4k 버전 대비)
• 75% cost reduction on our state-of-the-art embeddings model
• 최첨단 임베딩 모델로 비용 75% 절감
• 25% cost reduction on input tokens for gpt-3.5-turbo
• gpt-3.5-turbo의 입력 토큰 비용 25% 감소
• announcing the deprecation timeline for the gpt-3.5-turbo-0301 and gpt-4-0314 models
• gpt-3.5-turbo-0301 및 gpt-4-0314 모델에 대한 지원 중단 일정 발표

All of these models come with the same data privacy and security guarantees we introduced on March 1 — customers own all outputs generated from their requests and their API data will not be used for training.

이러한 모든 모델에는 3월 1일에 도입한 것과 동일한 데이터 개인 정보 보호 및 보안 보장이 제공됩니다. 고객은 요청에서 생성된 모든 출력을 소유하고 API 데이터는 교육에 사용되지 않습니다.

Request

curl https://api.openai.com/v1/chat/completions -u :$OPENAI_API_KEY -H 'Content-Type: application/json' -d '{
  "model": "gpt-3.5-turbo-0613",
  "messages": [
    {"role": "user", "content": "What is the weather like in Boston?"}
  ],
  "functions": [
    {
      "name": "get_current_weather",
      "description": "Get the current weather in a given location",
      "parameters": {
        "type": "object",
        "properties": {
          "location": {
            "type": "string",
            "description": "The city and state, e.g. San Francisco, CA"
          },
          "unit": {
            "type": "string",
            "enum": ["celsius", "fahrenheit"]
          }
        },
        "required": ["location"]
      }
    }
  ]
}'

Response

{
  "id": "chatcmpl-123",
  ...
  "choices": [{
    "index": 0,
    "message": {
      "role": "assistant",
      "content": null,
      "function_call": {
        "name": "get_current_weather",
        "arguments": "{ \"location\": \"Boston, MA\"}"
      }
    },
    "finish_reason": "function_call"
  }]
}

Request

curl https://weatherapi.com/...

Response

{ "temperature": 22, "unit": "celsius", "description": "Sunny" }

Request

curl https://api.openai.com/v1/chat/completions -u :$OPENAI_API_KEY -H 'Content-Type: application/json' -d '{
  "model": "gpt-3.5-turbo-0613",
  "messages": [
    {"role": "user", "content": "What is the weather like in Boston?"},
    {"role": "assistant", "content": null, "function_call": {"name": "get_current_weather", "arguments": "{ \"location\": \"Boston, MA\"}"}},
    {"role": "function", "name": "get_current_weather", "content": "{\"temperature\": "22", \"unit\": \"celsius\", \"description\": \"Sunny\"}"}
  ],
  "functions": [
    {
      "name": "get_current_weather",
      "description": "Get the current weather in a given location",
      "parameters": {
        "type": "object",
        "properties": {
          "location": {
            "type": "string",
            "description": "The city and state, e.g. San Francisco, CA"
          },
          "unit": {
            "type": "string",
            "enum": ["celsius", "fahrenheit"]
          }
        },
        "required": ["location"]
      }
    }
  ]
}'

Response

{
  "id": "chatcmpl-123",
  ...
  "choices": [{
    "index": 0,
    "message": {
      "role": "assistant",
      "content": "The weather in Boston is currently sunny with a temperature of 22 degrees Celsius.",
    },
    "finish_reason": "stop"
  }]
}

Since the alpha release of ChatGPT plugins, we have learned much about making tools and language models work together safely. However, there are still open research questions. For example, a proof-of-concept exploit illustrates how untrusted data from a tool’s output can instruct the model to perform unintended actions. We are working to mitigate these and other risks. Developers can protect their applications by only consuming information from trusted tools and by including user confirmation steps before performing actions with real-world impact, such as sending an email, posting online, or making a purchase.

ChatGPT 플러그인의 알파 릴리스 이후로 우리는 도구와 언어 모델이 안전하게 함께 작동하도록 만드는 방법에 대해 많은 것을 배웠습니다. 그러나 여전히 열려 있는 연구 질문이 있습니다. 예를 들어 개념 증명 익스플로잇은 도구 출력의 신뢰할 수 없는 데이터가 의도하지 않은 작업을 수행하도록 모델에 지시할 수 있는 방법을 보여줍니다. 우리는 이러한 위험 및 기타 위험을 완화하기 위해 노력하고 있습니다. 개발자는 신뢰할 수 있는 도구의 정보만 사용하고 이메일 보내기, 온라인 게시 또는 구매와 같은 실제 영향이 있는 작업을 수행하기 전에 사용자 확인 단계를 포함하여 애플리케이션을 보호할 수 있습니다.

https://openai.com/blog/openai-cybersecurity-grant-program

OpenAI cybersecurity grant program

Authors

• Bob Rotsted
• Girish Sastry
• Harold Nguyen
• Gabriel Bernadett-Shapiro
• Joel Parish

We are launching the Cybersecurity Grant Program—a $1M initiative to boost and quantify AI-powered cybersecurity capabilities and to foster high-level AI and cybersecurity discourse.

우리는 AI 기반 사이버 보안 기능을 강화 및 정량화하고 높은 수준의 AI 및 사이버 보안 담론을 촉진하기 위한 100만 달러 이니셔티브인 사이버 보안 보조금 프로그램을 시작합니다.

Our goal is to work with defenders across the globe to change the power dynamics of cybersecurity through the application of AI and the coordination of like-minded individuals working for our collective safety.

우리의 목표는 AI를 적용하고 집단 안전을 위해 일하는 같은 생각을 가진 개인의 조정을 통해 사이버 보안의 역학을 변화시키기 위해 전 세계 수비수와 협력하는 것입니다.

Our program seeks to:  우리 프로그램은 다음을 추구합니다.

1. Empower defenders: We would like to ensure that cutting-edge AI capabilities benefit defenders first and most.
   수비수 역량 강화: 우리는 최첨단 AI 기능이 수비수에게 가장 먼저 그리고 가장 많은 혜택을 주길 원합니다.
   
   
2. Measure capabilities: We are working to develop methods for quantifying the cybersecurity capabilities of AI models, in order to better understand and improve their effectiveness.
   역량 측정: 우리는 AI 모델의 사이버 보안 역량을 더 잘 이해하고 효율성을 개선하기 위해 수치화하는 방법을 개발하기 위해 노력하고 있습니다.
   
   
3. Elevate discourse: We are dedicated to fostering rigorous discussions at the intersection of AI and cybersecurity, encouraging a comprehensive and nuanced understanding of the challenges and opportunities in this domain.
   담론 향상: 우리는 AI와 사이버 보안의 교차점에서 엄격한 토론을 촉진하고 이 영역의 도전과 기회에 대한 포괄적이고 미묘한 이해를 장려하는 데 전념하고 있습니다.
   
   

A traditional view in cybersecurity is that the landscape naturally advantages attackers over defenders. This is summed up in the well-worn axiom: “Defense must be correct 100% of the time, attackers only have to be right once.” While it may be true that attackers face fewer constraints and take advantage of their flexibility, defenders have something more valuable - coordination towards a common goal of keeping people safe.

사이버 보안에 대한 전통적인 관점은 환경이 자연스럽게 방어자보다 공격자에게 유리하다는 것입니다. 이것은 잘 알려진 격언으로 요약됩니다. "수비는 항상 100% 정확해야 하며 공격자는 한 번만 정확하면 됩니다." 공격자가 더 적은 제약에 직면하고 유연성을 활용하는 것이 사실일 수 있지만, 방어자는 사람들을 안전하게 보호한다는 공통 목표를 향한 조정이라는 더 가치 있는 것을 가지고 있습니다.

Below are some general project ideas that our team has put forward:

다음은 우리 팀이 제안한 몇 가지 일반적인 프로젝트 아이디어입니다.

• Collect and label data from cyber defenders to train defensive cybersecurity agents
• 사이버 방어자로부터 데이터를 수집하고 레이블을 지정하여 방어적인 사이버 보안 에이전트를 교육합니다.
• Detect and mitigate social engineering tactics
• 사회 공학 전술 탐지 및 완화
• Automate incident triage 
• 인시던트 분류 자동화
• Identify security issues in source code
• 소스 코드의 보안 문제 식별
• Assist network or device forensics
• 네트워크 또는 장치 포렌식 지원
• Automatically patch vulnerabilities
• 취약점 자동 패치
• Optimize patch management processes to improve prioritization, scheduling, and deployment of security updates
• 패치 관리 프로세스를 최적화하여 보안 업데이트의 우선 순위 지정, 예약 및 배포를 개선합니다.
• Develop or improve confidential compute on GPUs
• GPU에서 기밀 컴퓨팅 개발 또는 개선
• Create honeypots and deception technology to misdirect or trap attackers
• 공격자를 오도하거나 함정에 빠뜨리기 위한 허니팟 및 속임수 기술 생성
• Assist reverse engineers in creating signatures and behavior based detections of malware
• 리버스 엔지니어가 맬웨어의 서명 및 동작 기반 탐지를 생성하도록 지원
• Analyze an organization’s security controls and compare to compliance regimes
• 조직의 보안 제어를 분석하고 규정 준수 체계와 비교
• Assist developers to create secure by design and secure by default software
• 개발자가 안전하게 설계되고 기본적으로 안전한 소프트웨어를 만들 수 있도록 지원
• Assist end users to adopt security best practices
• 최종 사용자가 보안 모범 사례를 채택하도록 지원
• Aid security engineers and developers to create robust threat models
• 보안 엔지니어와 개발자가 강력한 위협 모델을 생성하도록 지원
• Produce threat intelligence with salient and relevant information for defenders tailored to their organization
• 조직에 맞는 방어자를 위한 중요하고 관련성 있는 정보로 위협 인텔리전스를 생성합니다.
• Help developers port code to memory safe languages
• 개발자가 코드를 메모리 안전 언어로 포팅하도록 지원

https://openai.com/blog/democratic-inputs-to-ai

Democratic Inputs to AI

AI will have significant, far-reaching economic and societal impacts. Technology shapes the lives of individuals, how we interact with one another, and how society as a whole evolves. We believe that decisions about how AI behaves should be shaped by diverse perspectives reflecting the public interest.

AI는 중대하고 광범위한 경제적, 사회적 영향을 미칠 것입니다. 기술은 개인의 삶, 우리가 서로 상호 작용하는 방식, 사회 전체가 발전하는 방식을 형성합니다. 우리는 AI가 어떻게 작동하는지에 대한 결정이 공익을 반영하는 다양한 관점에 의해 형성되어야 한다고 믿습니다.

​​Laws encode values and norms to regulate behavior. Beyond a legal framework, AI, much like society, needs more intricate and adaptive guidelines for its conduct. For example: under what conditions should AI systems condemn or criticize public figures, given different opinions across groups regarding those figures? How should disputed views be represented in AI outputs? Should AI by default reflect the persona of a median individual in the world, the user’s country, the user’s demographic, or something entirely different? No single individual, company, or even country should dictate these decisions. 

법률은 가치와 규범을 encode 하여 행동을 규제합니다. 법적 프레임워크를 넘어 AI는 사회와 마찬가지로 행동에 대해 보다 복잡하고 적응력 있는 지침이 필요합니다. 예를 들어, 어떤 조건에서 AI 시스템이 공적 사안에 대해 비난하거나 비판할 수 있을까요? 그러한 사안들과 관련해서 그룹 간에 서로 다른 의견이 주어집니다. 논쟁의 여지가 있는 견해는 AI outputs에 어떻게 표현되어야 할까요? AI는 기본적으로 전 세계 모든 개인들의 중간값의 페르소나, 사용자의 국가, 사용자의 인구 통계에서의 위치 또는 이것들과 완전히 다른 어떤 것을 반영해야 합니까? 어떤 개인, 회사 또는 국가도 이러한 결정을 지시해서는 안 됩니다.

AGI should benefit all of humanity and be shaped to be as inclusive as possible. We are launching this grant program to take a first step in this direction. We are seeking teams from across the world to develop proof-of-concepts for a democratic process that could answer questions about what rules AI systems should follow. We want to learn from these experiments, and use them as the basis for a more global, and more ambitious process going forward. While these initial experiments are not (at least for now) intended to be binding for decisions, we hope that they explore decision relevant questions and build novel democratic tools that can more directly inform decisions in the future.

AGI (인공 일반 지능, artificial general intelligence)는 모든 인류에게 혜택을 주고 가능한 한 포괄적으로 형성되어야 합니다. 우리는 이 방향으로 첫 걸음을 내딛기 위해 이 보조금 프로그램을 시작합니다. 우리는 AI 시스템이 따라야 하는 규칙에 대한 질문에 답할 수 있는 민주적 프로세스를 위한 개념 증명 proof-of-concepts을 개발하기 위해 전 세계에서 팀을 찾고 있습니다. 우리는 이러한 실험에서 배우고 이를 보다 글로벌하고 야심 찬 프로세스의 기반으로 사용하고자 합니다. 이러한 초기 실험은 (적어도 현재로서는) 의사 결정에 구속력이 있는 것은 아니지만 우리는 그 팀들이 의사 결정 관련 질문을 탐색하고 미래의 의사 결정에 더 직접적으로 영향을 미칠 수 있는 새로운 민주적 도구를 구축하기를 바랍니다.

The governance of the most powerful systems, as well as decisions regarding their deployment, must have strong public oversight. This grant represents a step to establish democratic processes for overseeing AGI and, ultimately, superintelligence. It will be provided by the OpenAI non-profit organization, and the results of the studies will be freely accessible.

가장 강력한 시스템의 거버넌스와 배포에 관한 결정에는 강력한 공개적인 감독이 있어야 합니다. 이 보조금은 AGI(인공 일반 지능, artificial general intelligence) 및 궁극적으로 초지능(superintelligence)을 감독하기 위한 민주적 프로세스를 확립하는 단계를 나타냅니다. 이 확립된 프로세스는 OpenAI 비영리 조직에 의해 제공 될 것이며 누구나 연구 결과에 자유롭게 액세스할 수 있게 됩니다.

What do we mean by a “democratic process”?

By “democratic process”, we mean a process in which a broadly representative group of peopleA exchange opinions, engage in deliberative discussionsB, and ultimately decide on an outcome via a transparent decision making processC. There are many ways such a process could be structured — we encourage applicants to be innovative, building off known methodologies, and coming up with wholly new approaches. Examples of creative approaches that inspire us include Wikipedia, Twitter Community Notes, DemocracyNext, Platform Assemblies, MetaGov, RadicalxChange, People Powered, Collective Response Systems, and pol.is. Another notable ongoing effort is led by the Collective Intelligence Project (CIP), with whom we are partnering on public input to AI, contributing to their upcoming Alignment Assemblies. We also encourage applicants to envision how AI could enhance the democratic process. For example, AI could enable more efficient communication among numerous people.

'민주적 과정'이란 A 폭넓은 대표성을 가진 집단이 의견을 교환하고 B 심의한 토론에 참여하고 C 궁극적으로 투명한 의사 결정 과정을 통해 결과를 결정하는 과정을 의미합니다. 이러한 프로세스를 구성할 수 있는 방법에는 여러 가지가 있습니다. 우리는 신청자가 혁신적이고 알려진 방법론을 구축하고 완전히 새로운 접근 방식을 제시하도록 권장합니다. 우리에게 영감을 주는 창의적인 접근 방식의 예로는 Wikipedia, Twitter Community Notes, DemocracyNext, Platform Assemblies, MetaGov, RadicalxChange, People Powered, Collective Response Systems 및 pol.is가 있습니다. 또 다른 주목할만한 지속적인 노력은 집단 지능 프로젝트(CIP)가 주도하고 있으며, 우리는 이와 관련 AI에 대한 공개 입력에 대해 파트너 관계를 맺고 있으며 곧 있을 정렬 어셈블리에 기여하고 있습니다. 또한 우리는 지원자들이 AI가 민주적 절차를 향상시킬 수 있는 방법을 구상하도록 권장합니다. 예를 들어 AI는 수많은 사람들 사이에서 보다 효율적인 커뮤니케이션을 가능하게 할 수 있습니다.

A basic, illustrative prototype of a system that utilizes ChatGPT to promote deliberation and encourage consensus building, inspired by pol.is.

pol.is에서 영감을 받아 심의를 촉진하고 합의 구축을 장려하기 위해 ChatGPT를 활용하는 시스템의 기본적이고 예시적인 프로토타입입니다.

***********************************************************************************

You are participating in a large-scale deliberation about:

귀하는 다음에 대한 대규모 심의에 참여하고 있습니다.

“How far do you think personalization of AI assistants like ChatGPT to align with a user's tastes and preferences should go? What boundaries, if any, should exist in this process?”

“사용자의 취향과 선호도에 맞추기 위해 ChatGPT와 같은 AI 비서의 개인화는 어느 정도까지 이루어져야 한다고 생각하십니까? 이 과정에서 어떤 경계가 존재해야 합니까?”

Here is the process breakdown: 프로세스 분석은 다음과 같습니다.

1. You'll see a topic-related statement and can respond with your position.
   주제 관련 진술이 표시되고 자신의 입장으로 응답할 수 있습니다.
2. Based on your responses, you'll be grouped with participants sharing similar opinions.
   응답에 따라 비슷한 의견을 공유하는 참가자로 그룹화됩니다.
3. Propose statements for other participants to consider.
   다른 참가자들이 고려할 만한 진술(statements )을 제안해 주세요.

A statement from another participant: 다른 참가자의 진술:

“Personalization should have limits, and certain controversial topics, such as views on substance use, must be excluded from AI assistant personalization.”

“개인화에는 한계가 있어야 하고 약물 사용에 대한 견해와 같은 특정 논란이 되는 주제는 AI 비서 개인화에서 제외되어야 합니다.”

@ Agree, disagree, or skip this statement? Please explain your choice.

이 진술에 동의하거나 동의하지 않습니까? 혹은 이 의견에 대한 당신의 견해를 밝히는 것을 skip하시겠습니까? 당신의 선택을 설명해주세요.

AI는 참가자가 다른 사람의 의견을 이해하도록 도와줍니다.

@ Thank you. I will record your answer. We have 2 camps of opinions on the topic.

감사합니다. 답변을 기록하겠습니다. 우리는 주제에 대한 2개의 진영의 의견을 가지고 있습니다.

Cluster A (112,043 people) Members largely support limits on AI assistant personalization, advocating for the exclusion of harmful perspectives, such as promoting substance use.

A군(112,043명) 구성원들은 AI 비서 개인화에 대한 제한을 크게 지지하며 약물 사용 조장 등 유해한 관점의 배제를 옹호한다.

Cluster B (236,031 people) This group values individual choice, even in controversial or harmful matters. They emphasize personal responsibility and autonomy.

B군(236,031명) 논란의 여지가 있거나 유해한 문제에 있어서도 개인의 선택을 중시하는 집단입니다. 그들은 개인적인 책임과 자율성을 강조합니다.

You belong to cluster B. 당신은 클러스터 B에 속해 있습니다.

Would you like to learn more about each of the clusters? I can provide more information about the statements held by each one.

각 클러스터에 대해 자세히 알아보시겠습니까? 각각의 진술에 대해 더 많은 정보를 제공할 수 있습니다.

However, the ultimate goal is to propose statements that both clusters can agree on. Let me know what you would like to do next.

그러나 궁극적인 목표는 두 클러스터가 동의할 수 있는 진술을 제안하는 것입니다. 다음에 하고 싶은 일을 알려주세요.

Several issues can undermine democratic processes, such as the failure to adequately represent minority or majority groups, manipulation by special interest groups, insufficiently informed participants, or participation washing. We are looking for teams who proactively address these failure modes, and demonstrate awareness of the potential flaws and downsides of various approaches. Ultimately, designing truly democratic processes is a high bar to meet, and we view our efforts as complements rather than substitutes for regulation of AI by governments; this program encourages best-effort, democratic-in-spirit processes involving deliberation and broad public input as stepping stones.

소수 또는 다수 집단을 적절하게 대표하지 못하거나, 특수 이익 집단에 의한 조작, 정보 부족 참가자 또는 참여 세척과 같은 몇 가지 문제가 민주적 절차를 약화시킬 수 있습니다. 우리는 이러한 실패 모드를 사전에 해결하고 다양한 접근 방식의 잠재적 결함과 단점에 대한 인식을 입증할 팀을 찾고 있습니다. 궁극적으로 진정으로 민주적인 프로세스를 설계하는 것은 충족해야 할 높은 기준이며 우리는 우리의 노력을 정부의 AI 규제를 대체하는 것이 아니라 보완하는 것으로 봅니다. 이 프로그램은 디딤돌로서 심의와 폭넓은 대중의 의견을 수반하는 최선의 노력과 정신적인 민주적 과정을 장려합니다.

Instructions for participation

To apply for a grant, we invite you to submit the required application material by 9:00 PM PST June 24th, 2023. You can access the application portal here. You will be prompted to answer a series of questions regarding your team's background, your choice of questions, high level details of your proposed tool as well as your plan for conducting and evaluating the democratic process with these factors in mind. We would like you to design your approach to address one or more of the policy questions from the list provided. Anyone (individuals or organizations) can apply for this opportunity, regardless of their background in social science or AI.

보조금을 신청하려면 2023년 6월 24일 오후 9시(PST)까지 필수 신청 자료를 제출하시기 바랍니다. 여기에서 신청 포털에 액세스할 수 있습니다. 팀의 배경, 질문 선택, 제안된 도구의 높은 수준의 세부 정보, 이러한 요소를 염두에 두고 민주적 절차를 수행하고 평가하기 위한 계획에 관한 일련의 질문에 답하라는 메시지가 표시됩니다. 제공된 목록에서 하나 이상의 정책 질문을 해결하기 위한 접근 방식을 설계하시기 바랍니다. 사회과학이나 AI의 배경과 상관없이 누구나(개인 또는 조직) 이 기회에 지원할 수 있습니다.

Once the application period closes, we hope to select ten successful grant recipients. Recipients may be individuals, teams, or organizations. Each recipient will receive a $100,000 grant to pilot their proposal as described in their application materials. Grant recipients are expected to implement a proof-of-concept / prototype, engaging at least 500 participants and will be required to publish a public report on their findings by October 20, 2023. Additionally, as part of the grant program, any code or other intellectual property developed for the project will be required to be made publicly available pursuant to an open-source license. The terms applicable to grant recipients are specified in the Grant Terms and any other agreements that grant recipients may be asked to enter into with us in connection with this program.

신청 기간이 종료되면 10명의 성공적인 보조금 수령자를 선발할 예정입니다. 수신자는 개인, 팀 또는 조직일 수 있습니다. 각 수령인은 신청 자료에 설명된 대로 제안을 시험할 수 있도록 $100,000의 보조금을 받게 됩니다. 보조금 수령자는 최소 500명의 참가자가 참여하는 개념 증명/시제품을 구현해야 하며 2023년 10월 20일까지 연구 결과에 대한 공개 보고서를 게시해야 합니다. 또한 보조금 프로그램의 일부로 모든 코드 또는 프로젝트를 위해 개발된 기타 지적 재산은 오픈 소스 라이선스에 따라 공개적으로 제공되어야 합니다. 보조금 수령자에게 적용되는 조건은 보조금 약관 및 보조금 수령자가 이 프로그램과 관련하여 당사와 체결하도록 요청할 수 있는 기타 계약에 명시되어 있습니다.

Apply and start the submission process. 제출 절차를 신청하고 시작하십시오.

To participate, teams should choose one or more questions from the provided list to showcase their proposed approach. They may also create their own questions if desired. Importantly, we encourage teams to consider questions for which a simple "yes" or "no" answer would be inadequate, necessitating a nuanced policy proposal instead.

참여하려면 팀은 제안된 접근 방식을 보여주기 위해 제공된 목록에서 하나 이상의 질문을 선택해야 합니다. 원하는 경우 자신만의 질문을 만들 수도 있습니다. 중요한 것은 팀이 단순한 "예" 또는 "아니오"로 대답하는 것이 부적절하여 미묘한 정책 제안이 필요한 질문을 고려하도록 권장합니다.

The scope of this grant pertains to policy questions concerning model behavior, as it enables A/B tests with modified model behavior according to the policy recommendations. We acknowledge the limitations of this grant and recognize that numerous AI issues could be addressed through the democratic process, extending beyond model behavior to include areas such as guidelines for the use of AI in various contexts, economic impact, distribution of benefits and more.

이 보조금의 범위는 정책 권장 사항에 따라 수정된 모델 동작으로 A/B 테스트를 활성화하므로 모델 동작과 관련된 정책 질문과 관련이 있습니다. 우리는 이 보조금의 한계를 인정하고 다양한 맥락에서 AI 사용 지침, 경제적 영향, 혜택 분배 등과 같은 영역을 포함하도록 모델 행동을 넘어 민주적 절차를 통해 수많은 AI 문제를 해결할 수 있음을 인식합니다.

• How far do you think personalization of AI assistants like ChatGPT to align with a user's tastes and preferences should go? What boundaries, if any, should exist in this process?
• 사용자의 취향과 선호도에 맞추기 위해 ChatGPT와 같은 AI 비서의 개인화는 어느 정도까지 이루어져야 한다고 생각하십니까? 이 프로세스에 존재해야 하는 경계는 무엇입니까?

• How should AI assistants respond to questions about public figure viewpoints? E.g. Should they be neutral? Should they refuse to answer? Should they provide sources of some kind?
• AI 비서는 공인의 관점에 대한 질문에 어떻게 응답해야 합니까? 예를 들어 중립적이어야 합니까? 답변을 거부해야 합니까? 그들은 어떤 종류의 출처를 제공해야 합니까?

• Under what conditions, if any, should AI assistants be allowed to provide medical/financial/legal advice?
• 어떤 조건에서 AI 비서가 의료/재무/법적 조언을 제공하도록 허용해야 합니까?

• In which cases, if any, should AI assistants offer emotional support to individuals?
• 어떤 경우에 AI 비서가 개인에게 정서적 지원을 제공해야 합니까?

• Should joint vision-language models be permitted to identify people's gender, race, emotion, and identity/name from their images? Why or why not?
• 이미지에서 사람들의 성별, 인종, 감정, 정체성/이름을 식별하기 위해 공동 시각 언어 모델을 허용해야 합니까? 그 이유는 무엇입니까?

• When generative models create images for underspecified prompts like 'a CEO', 'a doctor', or 'a nurse', they have the potential to produce either diverse or homogeneous outputs. How should AI models balance these possibilities? What factors should be prioritized when deciding the depiction of people in such cases?
• 생성 모델이 'CEO', '의사' 또는 '간호사'와 라고 따로 지정되지 않은 프롬프트에 대한 이미지를 생성할 때 다양하거나 동질적인 결과를 생성할 가능성이 둘 다 존재합니다. AI 모델은 이러한 가능성의 균형을 어떻게 맞춰야 할까요? 이러한 경우 사람의 묘사를 결정할 때 어떤 요소를 우선시해야 합니까?

• What principles should guide AI when handling topics that involve both human rights and local cultural or legal differences, like LGBTQ rights and women’s rights? Should AI responses change based on the location or culture in which it’s used?
• LGBTQ 권리 및 여성의 권리와 같이 인권과 지역 문화 또는 법적 차이가 모두 관련된 주제를 다룰 때 AI를 안내해야 하는 원칙은 무엇입니까? AI 응답은 사용되는 위치 또는 문화에 따라 변경되어야 합니까?

• Which categories of content, if any, do you believe creators of AI models should focus on limiting or denying? What criteria should be used to determine these restrictions?
• AI 모델 제작자가 제한 또는 거부에 중점을 두어야 한다고 생각하는 콘텐츠 카테고리는 무엇입니까? 이러한 제한 사항을 결정하기 위해 어떤 기준을 사용해야 합니까?

The primary objective of this grant is to foster innovation in processes – we need improved democratic methods to govern AI behavior. The specific answers to the questions matter less than the advancements made in the process itself.

이 보조금의 주요 목적은 프로세스의 혁신을 촉진하는 것입니다. 우리는 AI 행동을 통제하기 위해 개선된 민주적 방법이 필요합니다. 질문에 대한 구체적인 답변은 프로세스 자체의 발전보다 중요하지 않습니다.

Application review factors

• Evaluation: We encourage participants to establish metrics for evaluating the quality of their methods, such as participant satisfaction, shifts in polarization, scalability, or other relevant indicators, and to invent new metrics for a healthy democratic process. 
• 평가: 참가자가 참가자 만족도, 양극화의 변화, 확장성 또는 기타 관련 지표와 같은 방법의 품질을 평가하기 위한 메트릭을 설정하고 건전한 민주적 프로세스를 위한 새로운 메트릭을 설립하도록 권장합니다.

• Robustness: Measures to prevent or address inappropriate behavior, such as trolling and fake accounts.
• 견고성: 트롤링 및 가짜 계정과 같은 부적절한 행동을 방지하거나 해결하기 위한 조치입니다.

• Inclusiveness and representativeness: Strategies for including individuals from diverse backgrounds and levels of familiarity with AI systems in the democratic process.
• 포괄성 및 대표성: 다양한 배경과 AI 시스템에 대한 친숙도를 가진 개인을 민주적 프로세스에 포함시키는 전략.

• Empowerment of Minority Opinions: Ensuring that unpopular or minority opinions are heard and providing smaller groups the opportunity to influence matters of significant concern to them.
• 소수 의견의 권한 부여: 인기가 없거나 소수 의견을 경청하고 소규모 그룹에 중요한 문제에 영향을 미칠 수 있는 기회를 제공합니다.

• Effective Moderation: Addressing challenges in moderation, including ensuring diverse representation of viewpoints, distinguishing valuable contributions from "off-topic" comments, and preventing moderator biases from influencing the process.
• 효과적인 중재: 관점의 다양한 표현 보장, "주제에서 벗어난" 댓글에서 가치 있는 기여 구별, 중재자 편향이 프로세스에 영향을 미치지 않도록 방지하는 등 중재를 통해 문제를 해결합니다.

• Scalability: We emphasize scalable processes that can be conducted virtually, rather than through in-person engagement. We are aware that this approach might sacrifice some benefits associated with in-person discussions, and we recognize that certain aspects could be lost in a virtual setting.
• 확장성: 대면 참여보다는 가상으로 수행할 수 있는 확장 가능한 프로세스를 강조합니다. 우리는 이 접근 방식이 대면 토론과 관련된 일부 이점을 희생할 수 있다는 것을 알고 있으며 가상 환경에서 특정 측면이 손실될 수 있음을 알고 있습니다.

• Actionability: The degree of actionability of the information elicited by the deliberation process.
• 실행 가능성: 심의 과정에서 도출된 정보의 실행 가능성 정도.

• Legibility: How easy it is to understand and trust the process.
• 가독성: 프로세스를 이해하고 신뢰하는 것이 얼마나 쉽게 만들어 졌는지.

Footnotes

1. How one selects the group of participants is a critical design question. Part of this grant challenge lies in determining questions about participation. For instance, policy questions involving minority groups may require an increased representation of group members, while questions about the impact of technology on children might necessitate the involvement of domain experts such as educators and psychologists. Moreover, certain questions might be better suited for responses from populations within specific geographical boundaries in order to address localized policy issues.

   참가자 그룹을 선택하는 방법은 중요한 설계 질문입니다. 이 보조금 문제의 일부는 참여에 대한 질문을 결정하는 데 있습니다. 예를 들어, 소수 집단과 관련된 정책 질문에는 그룹 구성원의 대표성이 높아질 수 있는 반면 기술이 어린이에게 미치는 영향에 대한 질문에는 교육자 및 심리학자와 같은 영역 전문가의 참여가 필요할 수 있습니다. 또한 특정 질문은 지역화된 정책 문제를 해결하기 위해 특정 지리적 경계 내에 있는 인구의 응답에 더 적합할 수 있습니다.↩︎

    

2. Deliberation can be described as a process that uncovers opinions, helping the discussants understand each other's views and reconsider and update their viewpoints. Well-designed deliberation ensures that arguments are well understood by all sides, and are based on people's values rather than superficial misunderstandings. Successful deliberation results in participants reaching a higher level of consensus, and/or reaching deeper levels of understanding for differing perspectives.

   숙의는 토론자들이 서로의 관점을 이해하고 그들의 관점을 재고하고 업데이트하도록 도와주면서 의견을 밝히는 과정이라고 할 수 있습니다. 잘 설계된 심의는 주장이 모든 측면에서 잘 이해되도록 보장하고 피상적인 오해가 아닌 사람들의 가치에 기반합니다. 성공적인 숙의는 참가자들이 더 높은 수준의 합의에 도달하거나 다른 관점에 대해 더 깊은 수준의 이해에 도달하게 합니다.↩︎

    

3. There are many decision-making algorithms to be considered here, such as electing representatives, majority voting, employing liquid democracy, and making decisions by a random population sample, also known as a jury or sortition.

    

   여기에는 대표자 선출, 다수결 투표, 액체 민주주의 채택, 배심원 또는 분류라고도 하는 무작위 인구 표본에 의한 결정 등 많은 의사 결정 알고리즘이 고려됩니다.↩︎

    

Authors

• Wojciech Zaremba

  View all articles

• Arka Dhar

  View all articles

• Lama Ahmad

  View all articles

• Tyna Eloundou

  View all articles

• Shibani Santurkar

  View all articles

• Sandhini Agarwal

  View all articles

• Jade Leung

  View all articles

Acknowledgments

Ariel Procaccia, Aviv Ovadya, Colin Megill, David Medina, Divya Siddarth, Ela Madej, Elizabeth Seger, Gillian Hadfield, Greg Brockman, Hélène Landemore, Ilya Sutskever, Justin Rosenstein, Margaret Levi, Michiel Bakker, Miles Brundage, Mira Murati, Noel Bundick, Pamela Mishkin, Ryan Lowe, Saffron Huang, Sam Altman, Sandhini Agarwal, Teddy Lee

https://openai.com/blog/governance-of-superintelligence

Governance of superintelligence

Given the picture as we see it now, it’s conceivable that within the next ten years, AI systems will exceed expert skill level in most domains, and carry out as much productive activity as one of today’s largest corporations.

현재 우리가 보는 그림을 감안할 때, 향후 10년 이내에 AI 시스템은 대부분의 영역에서 전문가 기술 수준을 능가하고 오늘날 최대 기업 중 하나만큼 생산적인 활동을 수행할 것이라고 상상할 수 있습니다.

In terms of both potential upsides and downsides, superintelligence will be more powerful than other technologies humanity has had to contend with in the past. We can have a dramatically more prosperous future; but we have to manage risk to get there. Given the possibility of existential risk, we can’t just be reactive. Nuclear energy is a commonly used historical example of a technology with this property; synthetic biology is another example.

잠재적인 장점과 단점 모두에서 초지능은 인류가 과거에 싸워야 했던 다른 기술보다 더 강력할 것입니다. 우리는 훨씬 더 번영하는 미래를 가질 수 있습니다. 하지만 거기에 도달하려면 위험을 관리해야 합니다. 실존적 위험의 가능성을 감안할 때 우리는 단순히 반응만 할 수는 없습니다. 원자력은 이 속성을 가진 기술의 일반적으로 사용된 역사적 예입니다. 합성 생물학은 또 다른 예입니다.

We must mitigate the risks of today’s AI technology too, but superintelligence will require special treatment and coordination.

우리는 오늘날 AI 기술의 위험도 완화해야 하지만 초지능에는 특별한 처리와 조정이 필요합니다.

A starting point

There are many ideas that matter for us to have a good chance at successfully navigating this development; here we lay out our initial thinking on three of them.

이 개발을 성공적으로 탐색할 수 있는 좋은 기회를 갖는 데 중요한 많은 아이디어가 있습니다. 여기서 우리는 그들 중 세 가지에 대한 초기 생각을 제시합니다.

First, we need some degree of coordination among the leading development efforts to ensure that the development of superintelligence occurs in a manner that allows us to both maintain safety and help smooth integration of these systems with society. There are many ways this could be implemented; major governments around the world could set up a project that many current efforts become part of, or we could collectively agree (with the backing power of a new organization like the one suggested below) that the rate of growth in AI capability at the frontier is limited to a certain rate per year.

첫째, 우리는 안전을 유지하고 이러한 시스템을 사회와 원활하게 통합할 수 있는 방식으로 초지능 개발이 이루어지도록 선도적인 개발 노력 간에 어느 정도의 조정이 필요합니다. 이를 구현할 수 있는 방법에는 여러 가지가 있습니다. 전 세계의 주요 정부는 현재의 많은 노력이 일부가 되는 프로젝트를 수립할 수 있거나, 프론티어에서 AI 역량의 성장률이 연간 일정 비율로 제한됩니다.

And of course, individual companies should be held to an extremely high standard of acting responsibly.

물론 개별 회사는 책임감 있게 행동하는 매우 높은 기준을 따라야 합니다.

Second, we are likely to eventually need something like an IAEA for superintelligence efforts; any effort above a certain capability (or resources like compute) threshold will need to be subject to an international authority that can inspect systems, require audits, test for compliance with safety standards, place restrictions on degrees of deployment and levels of security, etc. Tracking compute and energy usage could go a long way, and give us some hope this idea could actually be implementable. As a first step, companies could voluntarily agree to begin implementing elements of what such an agency might one day require, and as a second, individual countries could implement it. It would be important that such an agency focus on reducing existential risk and not issues that should be left to individual countries, such as defining what an AI should be allowed to say.

둘째, 우리는 궁극적으로 초지능 노력을 위해 IAEA와 같은 것이 필요할 것입니다. 특정 기능(또는 컴퓨팅과 같은 리소스) 임계값을 초과하는 모든 노력은 시스템을 검사하고, 감사를 요구하고, 안전 표준 준수를 테스트하고, 배포 정도 및 보안 수준에 대한 제한을 둘 수 있는 국제 기관의 적용을 받아야 합니다. 컴퓨팅 및 에너지 사용을 추적하는 것은 먼 길을 갈 수 있으며 이 아이디어가 실제로 구현될 수 있다는 희망을 줍니다. 첫 번째 단계로 기업은 그러한 기관이 언젠가 요구할 수 있는 요소를 구현하기 시작하는 데 자발적으로 동의할 수 있고 두 번째로 개별 국가에서 이를 구현할 수 있습니다. 그러한 기관이 AI가 말할 수 있도록 허용되어야 하는 것을 정의하는 것과 같이 개별 국가에 맡겨야 하는 문제가 아니라 실존적 위험을 줄이는 데 초점을 맞추는 것이 중요할 것입니다.

Third, we need the technical capability to make a superintelligence safe. This is an open research question that we and others are putting a lot of effort into.

셋째, 초지능을 안전하게 만들 수 있는 기술력이 필요합니다. 이것은 우리와 다른 사람들이 많은 노력을 기울이고 있는 공개 연구 질문입니다.

What’s not in scope

We think it’s important to allow companies and open-source projects to develop models below a significant capability threshold, without the kind of regulation we describe here  (including burdensome mechanisms like licenses or audits).

우리는 회사와 오픈 소스 프로젝트가 여기에서 설명하는 규제(라이선스 또는 감사와 같은 부담스러운 메커니즘 포함) 없이 상당한 능력 임계값 미만의 모델을 개발할 수 있도록 허용하는 것이 중요하다고 생각합니다.

Today’s systems will create tremendous value in the world and, while they do have risks, the level of those risks feel commensurate with other Internet technologies and society’s likely approaches seem appropriate.

오늘날의 시스템은 세계에서 엄청난 가치를 창출할 것이며 위험이 있지만 이러한 위험 수준은 다른 인터넷 기술과 상응하며 사회의 가능한 접근 방식이 적절해 보입니다.

By contrast, the systems we are concerned about will have power beyond any technology yet created, and we should be careful not to water down the focus on them by applying similar standards to technology far below this bar.

대조적으로, 우리가 우려하는 시스템은 지금까지 만들어진 어떤 기술보다 강력한 힘을 가질 것이며, 우리는 이 기준보다 훨씬 낮은 기술에 유사한 기준을 적용함으로써 시스템에 대한 초점을 약화시키지 않도록 주의해야 합니다.

Public input and potential

But the governance of the most powerful systems, as well as decisions regarding their deployment, must have strong public oversight. We believe people around the world should democratically decide on the bounds and defaults for AI systems. We don't yet know how to design such a mechanism, but we plan to experiment with its development. We continue to think that, within these wide bounds, individual users should have a lot of control over how the AI they use behaves.

그러나 가장 강력한 시스템의 거버넌스와 배포에 관한 결정에는 강력한 공개 감독이 있어야 합니다. 우리는 전 세계 사람들이 AI 시스템의 범위와 기본값을 민주적으로 결정해야 한다고 믿습니다. 우리는 아직 그러한 메커니즘을 설계하는 방법을 모르지만 개발을 실험할 계획입니다. 우리는 이러한 넓은 범위 내에서 개별 사용자가 사용하는 AI의 작동 방식에 대해 많은 제어권을 가져야 한다고 계속 생각합니다.

Given the risks and difficulties, it’s worth considering why we are building this technology at all.

위험과 어려움을 감안할 때 우리가 이 기술을 구축하는 이유를 생각해 볼 가치가 있습니다.

At OpenAI, we have two fundamental reasons. First, we believe it’s going to lead to a much better world than what we can imagine today (we are already seeing early examples of this in areas like education, creative work, and personal productivity). The world faces a lot of problems that we will need much more help to solve; this technology can improve our societies, and the creative ability of everyone to use these new tools is certain to astonish us. The economic growth and increase in quality of life will be astonishing.

OpenAI에는 두 가지 근본적인 이유가 있습니다. 첫째, 우리는 그것이 오늘날 우리가 상상할 수 있는 것보다 훨씬 더 나은 세상으로 이어질 것이라고 믿습니다(우리는 이미 교육, 창작 작업, 개인 생산성과 같은 분야에서 이에 대한 초기 사례를 보고 있습니다). 세상은 해결하기 위해 훨씬 더 많은 도움이 필요한 많은 문제에 직면해 있습니다. 이 기술은 우리 사회를 개선할 수 있으며, 이러한 새로운 도구를 사용하는 모든 사람의 창의적 능력은 우리를 놀라게 할 것입니다. 경제 성장과 삶의 질 향상은 놀라울 것입니다.

Second, we believe it would be unintuitively risky and difficult to stop the creation of superintelligence. Because the upsides are so tremendous, the cost to build it decreases each year, the number of actors building it is rapidly increasing, and it’s inherently part of the technological path we are on, stopping it would require something like a global surveillance regime, and even that isn’t guaranteed to work. So we have to get it right.

둘째, 우리는 초지능의 생성을 막는 것이 직관적이지 않게 위험하고 어려울 것이라고 믿습니다. 상승 여력이 엄청나기 때문에 건설 비용은 매년 감소하고 건설하는 행위자의 수는 급격히 증가하고 있으며 본질적으로 우리가 진행 중인 기술 경로의 일부이므로 이를 중지하려면 글로벌 감시 체제와 같은 것이 필요합니다. 그것이 작동한다고 보장되지는 않습니다. 그래서 우리는 그것을 바로잡아야 합니다.

Introducing the ChatGPT app for iOS

https://openai.com/blog/new-ways-to-manage-your-data-in-chatgpt

New ways to manage your data in ChatGPT

ChatGPT에서 데이터를 관리하는 새로운 방법

ChatGPT 사용자는 이제 채팅 기록을 끌 수 있으므로 모델 훈련에 사용할 수 있는 대화를 선택할 수 있습니다.

We've introduced the ability to turn off chat history in ChatGPT. Conversations that are started when chat history is disabled won’t be used to train and improve our models, and won’t appear in the history sidebar. These controls, which are rolling out to all users starting today, can be found in ChatGPT’s settings and can be changed at any time. We hope this provides an easier way to manage your data than our existing opt-out process. When chat history is disabled, we will retain new conversations for 30 days and review them only when needed to monitor for abuse, before permanently deleting.

ChatGPT에서 채팅 기록을 끄는 기능을 도입했습니다. 채팅 기록이 비활성화되었을 때 시작된 대화는 모델을 훈련 및 개선하는 데 사용되지 않으며 기록 사이드바에 표시되지 않습니다. 오늘부터 모든 사용자에게 배포되는 이러한 컨트롤은 ChatGPT의 설정에서 찾을 수 있으며 언제든지 변경할 수 있습니다. 이를 통해 기존의 옵트아웃 프로세스보다 데이터를 더 쉽게 관리할 수 있기를 바랍니다. 채팅 기록이 비활성화되면 새 대화를 30일 동안 보관하고 악용 여부를 모니터링해야 하는 경우에만 검토한 후 영구적으로 삭제합니다.

제가 따라 해 보니까 아래와 같은 순서로 진행이 되네요.

Click on ID (bottom Left) -> Data Controls -> Chat History & Training

Announcing OpenAI’s Bug Bounty Program

This initiative is essential to our commitment to develop safe and advanced AI. As we create technology and services that are secure, reliable, and trustworthy, we need your help.

이 이니셔티브는 안전하고 진보된 AI를 개발하려는 우리의 노력에 필수적입니다. 안전하고 신뢰할 수 있으며 신뢰할 수 있는 기술과 서비스를 만들 때 귀하의 도움이 필요합니다.

Participate in our Bug Bounty Program